GPT-5 is OpenAI's first all access model
OpenAI's GPT-5 is now available and is described as a "PhD-level expert," with deeper reasoning and new features for all ChatGPT users.
OpenAI’s GPT-5 is now official after several months of rumors, leaks, and teasers. The latest foundational model was unveiled in the brand’s summer update live stream on Thursday, where it showcased several demonstrations of the model’s prowess, across many disciplines, including coding, writing, language, and healthcare.
OpenAI CEO, Sam Altman, noted as the company worked on GPT-5, its primary service ChatGPT has grown to support 700 million weekly active users, up from 400 million monthly active users at last confirmation. The executive said that the development done to improve has made significant jumps in intelligence with each major version upgrade, indicating that GPT-3 was the equivalent of talking to a high school student, while GPT-4 was the equivalent of talking to a college student. He compared GPT-5 to having the assistance of a PHD level expert, in any area you need.
OpenAI calls GPT-5 its smartest and most efficient model to date with deeper reasoning ability and real‑time router function, which decides the complexity of a response based on the prompt. A user can ask the model to ‘think hard’ for a more detailed answer.
OpenAI has also unified GPT-5 under a single branded name, giving all tiers of users access to the model. The model is available as of Thursday. Free ChatGPT users will have access to GPT-5, and when they hit their limit, they will transition to GPT-5 mini, a smaller but still highly functional model within the GPT-5 family, which was compared to o3. Meanwhile, ChatGPT Plus users will have higher usage limits to GPT-5 than free users, and ChatGPT Pro users will have unlimited access to GPT-5, in addition to access to a model called GPT-5 Pro with additional features, such as Extended Thinking.
The GPT-5 default model will also be available for Team, Edu, and Enterprise plans, with generous usage limits for groups starting next week.
OpenAI showcased several interesting demos of how GPT-5 functions during its launch event including a code building a visual explanation of how airplanes stay in the air, and another code building a language learning app with interactive game features. To demonstrate how the model has improved in writing, the OpenAI team compared a eulogy GPT-5 generated sending off the former, now retired, models to one written by GPT-4o. The team highlighted that the GPT-5 had a punchier and more detailed writing style. To demonstrate its multimodal features, the OpenAI team showcased how users can converse with the model, with the study and learn mode to practice speaking a language in real time. With the recent memory feature, ChatGPT powered by GPT-5 can connect to Gmail and Google Calendar and help users build an easy-to-follow schedule with reminders and highlights of important tasks.
The OpenAI team also discussed the measures it has taken to improve the safety of GPT-5, in reducing hallucinations and deceptions. In particular, it has implemented a function called Safe Completions, which allows GPT-5 to give a limited answer to a question that might be challenging to answer for safety reasons. The model might also ask to follow up questions to see if the question requires a more advanced answer. If it determines it cannot answer the question, GPT-5 will give the user helpful alternatives.
Finally, Altman also spoke with an OpenAI team member who talked about using ChatGPT to understand the jargon of a complex medical diagnosis she found herself facing. She detailed that while older models helped her make critical decisions about her care and advocate for herself with doctors, GPT-5 provided more anticipatory information about why she was asking questions pertaining to a diagnosis. She and her partner also noted that the added benefit is that GPT-5 is readily available for people who may be now receiving a diagnosis that they need to navigate.